Testing Interactions in Classification Problems

Abstract

The purpose is to demonstrate a procedure for testing the increment to classification accuracy afforded by an interaction term using predictive discriminant analysis (PDA) and logistic regression (LR). The PSAT (Math) and ethnicity were employed to predict students at risk of passing or failing the Florida Comprehensive Assessment Test (FCAT) Math. Results favored PDA over LR for the failure group. For PDA, at an a priori alpha of .05, the moderation of ethnicity was not significant for the failure group, but was for both passing and the total sample, as were hit-rates for total sample and separate groups by Ethnicity.